The Bloody Doctor and Possible Implications for Revenue Managers

Should United Airlines CEO Oscar Muñoz ever use the term ‘customer-centricity’ again, and should someone then suggest he wash his mouth with soap, I would echo that. Thanks but no thanks to United for giving algorithmic decision making a bad name…

A lot has been said and written already about the case of the doctor who was forcibly removed from an overbooked flight, but at the end of the day, overbooking is Revenue Management, and that makes this (also) an RM nightmare. It seems fair to ask then what the RM community should make of this. If we look beyond the shameful customer service, is there an interesting case study, and lessons to be learned for RM practitioners?

This sad case illustrates very effectively how extremely price-insensitive passengers become very close to departure: a whole plane-load of passengers could not be enticed to shift their demand to the next flight for a considerable financial reward. Let’s try to quantify a few things to start with, using publicly available information.

UAL3411 on April 9 was carried out with an Embraer 170; it took 80 minutes to cover the 483kms between Chicago and Louisville, departing two hours behind schedule at 7:40pm (the doctor put up a good fight!). With 70 seats, and 4 people randomly picked for off-loading, the incentive must have been offered unsuccessfully to at least 74 people. Before boarding, the incentive offered was $400, in travel vouchers. It was increased to $800 and then $1,000 after boarding, but to no avail. When the carrot didn’t work the stick did the trick, at least for the unfortunate doctor in seat 12D, which is, by the way, a standard Economy Class seat that is ‘missing a window’, according to Seatguru (no Dr. Watson, let’s not digress...)[1].

As the destination is not a United hub, few passengers would have missed a connection by taking a later flight. Between these two airports, there are 8 direct flights on a Sunday, two departing after UAL3411, and one also operated by United; its relative pricing suggests that it is typically less busy. However, let’s give United some credit and assume their own employees would have taken the later flight if it had any seats available. And let’s assume that the American Airlines flight at 6:40pm was not a viable option either. In that case, the ‘shift of demand’ would have meant a (free) overnight stay at an airport hotel, catching the 7:35am flight the next day and arriving in Louisville just before 10am. Quite inconvenient for most, but surely surmountable for some?

Generally, price elasticities for short-haul air travel are assumed or estimated to be somewhere between -0.7 and -1.5. At the price-insensitive end, that means that raising the price by 10% would reduce demand by 7%. In this case, one could argue that offering $1,000 not to travel is equivalent to increasing the price by $1,000, as $1,000 is now the opportunity cost for those not taking up the offer. As vouchers are not as valuable as cash, let’s work with a cash equivalent value of $500. And while many passengers would have connected from other origins, let’s take flightaware’s reported median fare of $180.50 for the Chicago to Louisville market to benchmark the incentive against, i.e. consider the $500 cash equivalent incentive as a 277% ‘price’ increase. And if we ignore the integer nature of the problem at this level, the resulting decrease in demand was less than 0.5 out of 74 passengers, so less than 0.7%. That means that demand would be less elastic than -0.002, or a whopping 300 times less price-sensitive than the ‘typical’ case of -0.7 price elasticity.

These numbers raise a variety of serious questions, for RM, CRM and commercial airline management in general[2] 

  1. This flight must be a bit of a gem; the refusal to move suggests a product loyalty that would make every marketeer drool. Or not… But seriously, the combined compensation for the four overbooked passengers on this flight alone could have saved a life (I’m not making this up, it actually costs $3,337.06). What went wrong with 74 passengers’ utility function? Were they all due in hospital the next morning? Did they just switch off completely (not a bad approach on the average commuter flight)? Did they not want to ‘negotiate’ in public? I suspect that economics, statistics and data analytics may need guidance from psychology and perhaps even sociology to explain this.
  2. Shouldn’t the wealth of CRM data, along with the crew’s experience, enable United to pick someone other than a 69 year old doctor for such treatment? The numbers suggest it would be worthwhile: getting it wrong in this one case –  only 0.03% of United’s annual involuntary denied boardings[3]– temporarily wiped $1 Bn off UAL’s market value. At these extremes, a tweak to the overbooking algorithm won’t suffice; overbooking as the result of a rational balancing of expected benefits and costs does not work for the very small proportion of cases with a very huge reputational impact.
  3. It also shows again that price elasticity – or presumably any other measure of price-sensitivity – isn’t stable through time. It is probably safe to assume that at a slightly earlier stage (before making the boarding call, or just before passengers would have come to the airport), demand could have been shifted by much smaller incentives. So when exactly, and why, does it change so drastically? Why does it happen more strongly for some flights than for others, with very similar features and demand? What are the implications, not just for dynamic pricing, but for strategic price profiling too? And finally, if we don’t really know the answer to these questions, how comfortable are we with the use of price-sensitivity parameters in our RM systems?
  4. The Revenue Manager’s kneejerk reaction to price-insensitive demand is to put the price up. Fair enough, but in this particular case, when the objective is actually to shift demand by targeting relatively price-sensitive passengers, the implication is that denied boarding negotiation should never make it to the gate, let alone into the plane. The same people who wouldn’t move an inch (voluntarily…) for $1,000 once boarded, may have happily given up their seat only an hour before. Pro-actively identifying those people before they arrive at the gate is a challenge, but passenger profiling can guide this effort, and making personalised offers discretely can further increase the conversion rate.
  5. Alternatively, or additionally, it could be argued that this case re-confirms that ‘price’ sensitivity differs greatly across the various components that make up ‘the price’. Already, we allowed for a 50% reduction off the face value of the voucher to get to a cash value. For some passengers, e.g. frequent flyers who pay for their own travel, such a discount seems irrationally large and yet, 50% could well overestimate the perceived value of vouchers. Offering vouchers instead of cash make sense from a financial perspective, but what causes this low perceived value? Would there be more effective currencies (other than cash), such as points, future upgrades, lounge access vouchers, status upgrades, or perhaps a case of the best wine that $540 can buy…?
  6. We already knew that RM doesn’t work ‘on average(s)’, but has to be applied at a micro-level (‘the right price for the right passenger’ etc.). But what about the data inputs? In a presumably more normal situation, where four passengers would have taken the $400 vouchers, equivalent to say $200 in cash, demand would statistically look a factor 20 less price-sensitive. One family without any pressing plans for the next day could make the difference; does that feel like a robust basis for applying any kind of logic based on (historical) averages? And if it doesn’t for overbooking, what are the implications for RM at large?

Hopefully, the pictures of the poor doctor being thrown out of the plane will not be a precursor to the overbooking baby being thrown out with the muddy waters this sorry episode has created. For that is clearly not in anyone’s interest. But it would be good if a better understanding of individual passengers can mitigate the clear and stubborn shortcomings of average-based decision-making. In the press, success would be harder to detect than failure, but sure enough, things that do not happen can be progress too!

[1] The video suggests the doctor may actually have been in seat 17D, which has its window merely ‘misaligned’, but I promise this is the only poetic license taken in writing this.

[2] In line with consulting best practice, this opinion(ated) paper focuses on raising questions, rather than offering meaningful answers.